pith. sign in

arxiv: 2605.27139 · v1 · pith:E5ZJC26Tnew · submitted 2026-05-26 · 📡 eess.IV · cs.CV· physics.ins-det

Unsupervised Deep Image Prior for Sparse-View and Limited-Angle Electron Tomography

Serge Brosset , Daniel del Pozo Bueno , Thomas David , Laure Guetaz , Philippe Ciuciu , Zineb Saghi This is my paper

Pith reviewed 2026-06-29 15:08 UTC · model grok-4.3

classification 📡 eess.IV cs.CVphysics.ins-det
keywords datadeeplimited-anglesparse-viewtomographyconditionsdegradedelectron
0
0 comments X

The pith

Unsupervised deep image prior yields 3D reconstructions in electron tomography comparable to supervised methods for 60° tilt ranges and 10° increments on simulated data and enables reliable quantification on experimental data.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Electron tomography builds 3D pictures of materials by combining many 2D projections taken at different angles. When the range of angles is small or the number of projections is low, standard reconstruction methods produce blurry or artifact-filled results that are hard to interpret. The authors test a technique known as deep image prior, which uses a neural network without any pre-training on example images. The network structure itself acts as a regularizer to find a plausible 3D volume that matches the available projections. On computer-generated test cases, the method performed similarly to supervised deep learning approaches that require large training datasets, even when the tilt range was restricted to 60 degrees with 10-degree steps. When applied to real experimental datasets, the reconstructions supported quantitative 3D measurements that would otherwise be unreliable. The approach is presented as potentially useful across different materials and imaging setups where complete data collection is impractical.

Core claim

its performance is comparable to that of supervised approaches requiring training datasets, even for tilt ranges as limited as 60° and tilt increments of 10°.

Load-bearing premise

That the deep image prior network can be applied directly to highly degraded electron tomography acquisitions and produce quantitatively reliable 3D results without domain-specific modifications or additional constraints.

read the original abstract

Electron tomography (ET) plays an important role in the three-dimensional (3D) characterization of nanomaterials. However, under limited-angle and sparse-view conditions, conventional algorithms produce degraded reconstructions, which compromise the quality and interpretability of resulting 3D data. In this paper, we present deep image prior (DIP), an unsupervised deep learning (DL) approach, for highly degraded tomography acquisitions and demonstrate, using simulated data, that its performance is comparable to that of supervised approaches requiring training datasets, even for tilt ranges as limited as 60{\deg} and tilt increments of 10{\deg}. We then apply it to experimental data and show that it enables reliable 3D quantification under both sparse-view and limited-angle conditions, highlighting its potential for a wide range of materials and acquisition modalities.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no equations, parameters, or explicit assumptions beyond naming the method; ledger is therefore empty.

pith-pipeline@v0.9.1-grok · 5688 in / 1045 out tokens · 36238 ms · 2026-06-29T15:08:59.280600+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Unsupervised Deep Learning for Limited-Angle STEM-EDX Tomography -- Application to 3D Chemical Analysis of Phase-Change Memory Devices

    eess.IV 2026-06 unverdicted novelty 5.0

    Unsupervised multi-channel DIP-TV reconstructs near-isotropic 3D elemental maps from limited-angle EDX tomography data using only EDX signals, applied to GST memory devices in virgin and SET states.